Can Symptom Search Trends Forecast Respiratory Virus Patient Counts? A Rhinovirus Case Study in Korean Syndromic Surveillance

Official respiratory-virus counts by governments are reliable but often delayed, while symptom-related search activity can appear before updated surveillance reports. We ask whether respiratory-virus patient counts can be forecast from online Relative Search Volume (RSV), a platform-normalized search-interest index, without using past patient counts. Using 270 weeks of Korea Disease Control and Prevention Agency (KDCA) surveillance data, we construct a structured dataset aligning patient counts with Naver and Google RSV for symptom queries expanded through an LLM-assisted pipeline. In a Rhinovirus case study, existing time-series backbones trained with patient counts but restricted to search-history inputs at prediction time produce RSV-only forecasts that partially track patient trajectories and achieve performance comparable to patient-history baselines.